aboutsummaryrefslogtreecommitdiffstats
path: root/mm/page_alloc.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
commit1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch)
tree0bba044c4ce775e45a88a51686b5d9f90697ea9d /mm/page_alloc.c
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
Diffstat (limited to 'mm/page_alloc.c')
-rw-r--r--mm/page_alloc.c2220
1 files changed, 2220 insertions, 0 deletions
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
new file mode 100644
index 000000000000..c73dbbc1cd8f
--- /dev/null
+++ b/mm/page_alloc.c
@@ -0,0 +1,2220 @@
1/*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
16
17#include <linux/config.h>
18#include <linux/stddef.h>
19#include <linux/mm.h>
20#include <linux/swap.h>
21#include <linux/interrupt.h>
22#include <linux/pagemap.h>
23#include <linux/bootmem.h>
24#include <linux/compiler.h>
25#include <linux/module.h>
26#include <linux/suspend.h>
27#include <linux/pagevec.h>
28#include <linux/blkdev.h>
29#include <linux/slab.h>
30#include <linux/notifier.h>
31#include <linux/topology.h>
32#include <linux/sysctl.h>
33#include <linux/cpu.h>
34#include <linux/cpuset.h>
35#include <linux/nodemask.h>
36#include <linux/vmalloc.h>
37
38#include <asm/tlbflush.h>
39#include "internal.h"
40
41/*
42 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
43 * initializer cleaner
44 */
45nodemask_t node_online_map = { { [0] = 1UL } };
46nodemask_t node_possible_map = NODE_MASK_ALL;
47struct pglist_data *pgdat_list;
48unsigned long totalram_pages;
49unsigned long totalhigh_pages;
50long nr_swap_pages;
51
52/*
53 * results with 256, 32 in the lowmem_reserve sysctl:
54 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
55 * 1G machine -> (16M dma, 784M normal, 224M high)
56 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
57 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
58 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
59 */
60int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 32 };
61
62EXPORT_SYMBOL(totalram_pages);
63EXPORT_SYMBOL(nr_swap_pages);
64
65/*
66 * Used by page_zone() to look up the address of the struct zone whose
67 * id is encoded in the upper bits of page->flags
68 */
69struct zone *zone_table[1 << (ZONES_SHIFT + NODES_SHIFT)];
70EXPORT_SYMBOL(zone_table);
71
72static char *zone_names[MAX_NR_ZONES] = { "DMA", "Normal", "HighMem" };
73int min_free_kbytes = 1024;
74
75unsigned long __initdata nr_kernel_pages;
76unsigned long __initdata nr_all_pages;
77
78/*
79 * Temporary debugging check for pages not lying within a given zone.
80 */
81static int bad_range(struct zone *zone, struct page *page)
82{
83 if (page_to_pfn(page) >= zone->zone_start_pfn + zone->spanned_pages)
84 return 1;
85 if (page_to_pfn(page) < zone->zone_start_pfn)
86 return 1;
87#ifdef CONFIG_HOLES_IN_ZONE
88 if (!pfn_valid(page_to_pfn(page)))
89 return 1;
90#endif
91 if (zone != page_zone(page))
92 return 1;
93 return 0;
94}
95
96static void bad_page(const char *function, struct page *page)
97{
98 printk(KERN_EMERG "Bad page state at %s (in process '%s', page %p)\n",
99 function, current->comm, page);
100 printk(KERN_EMERG "flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
101 (int)(2*sizeof(page_flags_t)), (unsigned long)page->flags,
102 page->mapping, page_mapcount(page), page_count(page));
103 printk(KERN_EMERG "Backtrace:\n");
104 dump_stack();
105 printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n");
106 page->flags &= ~(1 << PG_private |
107 1 << PG_locked |
108 1 << PG_lru |
109 1 << PG_active |
110 1 << PG_dirty |
111 1 << PG_swapcache |
112 1 << PG_writeback);
113 set_page_count(page, 0);
114 reset_page_mapcount(page);
115 page->mapping = NULL;
116 tainted |= TAINT_BAD_PAGE;
117}
118
119#ifndef CONFIG_HUGETLB_PAGE
120#define prep_compound_page(page, order) do { } while (0)
121#define destroy_compound_page(page, order) do { } while (0)
122#else
123/*
124 * Higher-order pages are called "compound pages". They are structured thusly:
125 *
126 * The first PAGE_SIZE page is called the "head page".
127 *
128 * The remaining PAGE_SIZE pages are called "tail pages".
129 *
130 * All pages have PG_compound set. All pages have their ->private pointing at
131 * the head page (even the head page has this).
132 *
133 * The first tail page's ->mapping, if non-zero, holds the address of the
134 * compound page's put_page() function.
135 *
136 * The order of the allocation is stored in the first tail page's ->index
137 * This is only for debug at present. This usage means that zero-order pages
138 * may not be compound.
139 */
140static void prep_compound_page(struct page *page, unsigned long order)
141{
142 int i;
143 int nr_pages = 1 << order;
144
145 page[1].mapping = NULL;
146 page[1].index = order;
147 for (i = 0; i < nr_pages; i++) {
148 struct page *p = page + i;
149
150 SetPageCompound(p);
151 p->private = (unsigned long)page;
152 }
153}
154
155static void destroy_compound_page(struct page *page, unsigned long order)
156{
157 int i;
158 int nr_pages = 1 << order;
159
160 if (!PageCompound(page))
161 return;
162
163 if (page[1].index != order)
164 bad_page(__FUNCTION__, page);
165
166 for (i = 0; i < nr_pages; i++) {
167 struct page *p = page + i;
168
169 if (!PageCompound(p))
170 bad_page(__FUNCTION__, page);
171 if (p->private != (unsigned long)page)
172 bad_page(__FUNCTION__, page);
173 ClearPageCompound(p);
174 }
175}
176#endif /* CONFIG_HUGETLB_PAGE */
177
178/*
179 * function for dealing with page's order in buddy system.
180 * zone->lock is already acquired when we use these.
181 * So, we don't need atomic page->flags operations here.
182 */
183static inline unsigned long page_order(struct page *page) {
184 return page->private;
185}
186
187static inline void set_page_order(struct page *page, int order) {
188 page->private = order;
189 __SetPagePrivate(page);
190}
191
192static inline void rmv_page_order(struct page *page)
193{
194 __ClearPagePrivate(page);
195 page->private = 0;
196}
197
198/*
199 * Locate the struct page for both the matching buddy in our
200 * pair (buddy1) and the combined O(n+1) page they form (page).
201 *
202 * 1) Any buddy B1 will have an order O twin B2 which satisfies
203 * the following equation:
204 * B2 = B1 ^ (1 << O)
205 * For example, if the starting buddy (buddy2) is #8 its order
206 * 1 buddy is #10:
207 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
208 *
209 * 2) Any buddy B will have an order O+1 parent P which
210 * satisfies the following equation:
211 * P = B & ~(1 << O)
212 *
213 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
214 */
215static inline struct page *
216__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
217{
218 unsigned long buddy_idx = page_idx ^ (1 << order);
219
220 return page + (buddy_idx - page_idx);
221}
222
223static inline unsigned long
224__find_combined_index(unsigned long page_idx, unsigned int order)
225{
226 return (page_idx & ~(1 << order));
227}
228
229/*
230 * This function checks whether a page is free && is the buddy
231 * we can do coalesce a page and its buddy if
232 * (a) the buddy is free &&
233 * (b) the buddy is on the buddy system &&
234 * (c) a page and its buddy have the same order.
235 * for recording page's order, we use page->private and PG_private.
236 *
237 */
238static inline int page_is_buddy(struct page *page, int order)
239{
240 if (PagePrivate(page) &&
241 (page_order(page) == order) &&
242 !PageReserved(page) &&
243 page_count(page) == 0)
244 return 1;
245 return 0;
246}
247
248/*
249 * Freeing function for a buddy system allocator.
250 *
251 * The concept of a buddy system is to maintain direct-mapped table
252 * (containing bit values) for memory blocks of various "orders".
253 * The bottom level table contains the map for the smallest allocatable
254 * units of memory (here, pages), and each level above it describes
255 * pairs of units from the levels below, hence, "buddies".
256 * At a high level, all that happens here is marking the table entry
257 * at the bottom level available, and propagating the changes upward
258 * as necessary, plus some accounting needed to play nicely with other
259 * parts of the VM system.
260 * At each level, we keep a list of pages, which are heads of continuous
261 * free pages of length of (1 << order) and marked with PG_Private.Page's
262 * order is recorded in page->private field.
263 * So when we are allocating or freeing one, we can derive the state of the
264 * other. That is, if we allocate a small block, and both were
265 * free, the remainder of the region must be split into blocks.
266 * If a block is freed, and its buddy is also free, then this
267 * triggers coalescing into a block of larger size.
268 *
269 * -- wli
270 */
271
272static inline void __free_pages_bulk (struct page *page,
273 struct zone *zone, unsigned int order)
274{
275 unsigned long page_idx;
276 int order_size = 1 << order;
277
278 if (unlikely(order))
279 destroy_compound_page(page, order);
280
281 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
282
283 BUG_ON(page_idx & (order_size - 1));
284 BUG_ON(bad_range(zone, page));
285
286 zone->free_pages += order_size;
287 while (order < MAX_ORDER-1) {
288 unsigned long combined_idx;
289 struct free_area *area;
290 struct page *buddy;
291
292 combined_idx = __find_combined_index(page_idx, order);
293 buddy = __page_find_buddy(page, page_idx, order);
294
295 if (bad_range(zone, buddy))
296 break;
297 if (!page_is_buddy(buddy, order))
298 break; /* Move the buddy up one level. */
299 list_del(&buddy->lru);
300 area = zone->free_area + order;
301 area->nr_free--;
302 rmv_page_order(buddy);
303 page = page + (combined_idx - page_idx);
304 page_idx = combined_idx;
305 order++;
306 }
307 set_page_order(page, order);
308 list_add(&page->lru, &zone->free_area[order].free_list);
309 zone->free_area[order].nr_free++;
310}
311
312static inline void free_pages_check(const char *function, struct page *page)
313{
314 if ( page_mapcount(page) ||
315 page->mapping != NULL ||
316 page_count(page) != 0 ||
317 (page->flags & (
318 1 << PG_lru |
319 1 << PG_private |
320 1 << PG_locked |
321 1 << PG_active |
322 1 << PG_reclaim |
323 1 << PG_slab |
324 1 << PG_swapcache |
325 1 << PG_writeback )))
326 bad_page(function, page);
327 if (PageDirty(page))
328 ClearPageDirty(page);
329}
330
331/*
332 * Frees a list of pages.
333 * Assumes all pages on list are in same zone, and of same order.
334 * count is the number of pages to free, or 0 for all on the list.
335 *
336 * If the zone was previously in an "all pages pinned" state then look to
337 * see if this freeing clears that state.
338 *
339 * And clear the zone's pages_scanned counter, to hold off the "all pages are
340 * pinned" detection logic.
341 */
342static int
343free_pages_bulk(struct zone *zone, int count,
344 struct list_head *list, unsigned int order)
345{
346 unsigned long flags;
347 struct page *page = NULL;
348 int ret = 0;
349
350 spin_lock_irqsave(&zone->lock, flags);
351 zone->all_unreclaimable = 0;
352 zone->pages_scanned = 0;
353 while (!list_empty(list) && count--) {
354 page = list_entry(list->prev, struct page, lru);
355 /* have to delete it as __free_pages_bulk list manipulates */
356 list_del(&page->lru);
357 __free_pages_bulk(page, zone, order);
358 ret++;
359 }
360 spin_unlock_irqrestore(&zone->lock, flags);
361 return ret;
362}
363
364void __free_pages_ok(struct page *page, unsigned int order)
365{
366 LIST_HEAD(list);
367 int i;
368
369 arch_free_page(page, order);
370
371 mod_page_state(pgfree, 1 << order);
372
373#ifndef CONFIG_MMU
374 if (order > 0)
375 for (i = 1 ; i < (1 << order) ; ++i)
376 __put_page(page + i);
377#endif
378
379 for (i = 0 ; i < (1 << order) ; ++i)
380 free_pages_check(__FUNCTION__, page + i);
381 list_add(&page->lru, &list);
382 kernel_map_pages(page, 1<<order, 0);
383 free_pages_bulk(page_zone(page), 1, &list, order);
384}
385
386
387/*
388 * The order of subdivision here is critical for the IO subsystem.
389 * Please do not alter this order without good reasons and regression
390 * testing. Specifically, as large blocks of memory are subdivided,
391 * the order in which smaller blocks are delivered depends on the order
392 * they're subdivided in this function. This is the primary factor
393 * influencing the order in which pages are delivered to the IO
394 * subsystem according to empirical testing, and this is also justified
395 * by considering the behavior of a buddy system containing a single
396 * large block of memory acted on by a series of small allocations.
397 * This behavior is a critical factor in sglist merging's success.
398 *
399 * -- wli
400 */
401static inline struct page *
402expand(struct zone *zone, struct page *page,
403 int low, int high, struct free_area *area)
404{
405 unsigned long size = 1 << high;
406
407 while (high > low) {
408 area--;
409 high--;
410 size >>= 1;
411 BUG_ON(bad_range(zone, &page[size]));
412 list_add(&page[size].lru, &area->free_list);
413 area->nr_free++;
414 set_page_order(&page[size], high);
415 }
416 return page;
417}
418
419void set_page_refs(struct page *page, int order)
420{
421#ifdef CONFIG_MMU
422 set_page_count(page, 1);
423#else
424 int i;
425
426 /*
427 * We need to reference all the pages for this order, otherwise if
428 * anyone accesses one of the pages with (get/put) it will be freed.
429 * - eg: access_process_vm()
430 */
431 for (i = 0; i < (1 << order); i++)
432 set_page_count(page + i, 1);
433#endif /* CONFIG_MMU */
434}
435
436/*
437 * This page is about to be returned from the page allocator
438 */
439static void prep_new_page(struct page *page, int order)
440{
441 if (page->mapping || page_mapcount(page) ||
442 (page->flags & (
443 1 << PG_private |
444 1 << PG_locked |
445 1 << PG_lru |
446 1 << PG_active |
447 1 << PG_dirty |
448 1 << PG_reclaim |
449 1 << PG_swapcache |
450 1 << PG_writeback )))
451 bad_page(__FUNCTION__, page);
452
453 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
454 1 << PG_referenced | 1 << PG_arch_1 |
455 1 << PG_checked | 1 << PG_mappedtodisk);
456 page->private = 0;
457 set_page_refs(page, order);
458 kernel_map_pages(page, 1 << order, 1);
459}
460
461/*
462 * Do the hard work of removing an element from the buddy allocator.
463 * Call me with the zone->lock already held.
464 */
465static struct page *__rmqueue(struct zone *zone, unsigned int order)
466{
467 struct free_area * area;
468 unsigned int current_order;
469 struct page *page;
470
471 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
472 area = zone->free_area + current_order;
473 if (list_empty(&area->free_list))
474 continue;
475
476 page = list_entry(area->free_list.next, struct page, lru);
477 list_del(&page->lru);
478 rmv_page_order(page);
479 area->nr_free--;
480 zone->free_pages -= 1UL << order;
481 return expand(zone, page, order, current_order, area);
482 }
483
484 return NULL;
485}
486
487/*
488 * Obtain a specified number of elements from the buddy allocator, all under
489 * a single hold of the lock, for efficiency. Add them to the supplied list.
490 * Returns the number of new pages which were placed at *list.
491 */
492static int rmqueue_bulk(struct zone *zone, unsigned int order,
493 unsigned long count, struct list_head *list)
494{
495 unsigned long flags;
496 int i;
497 int allocated = 0;
498 struct page *page;
499
500 spin_lock_irqsave(&zone->lock, flags);
501 for (i = 0; i < count; ++i) {
502 page = __rmqueue(zone, order);
503 if (page == NULL)
504 break;
505 allocated++;
506 list_add_tail(&page->lru, list);
507 }
508 spin_unlock_irqrestore(&zone->lock, flags);
509 return allocated;
510}
511
512#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
513static void __drain_pages(unsigned int cpu)
514{
515 struct zone *zone;
516 int i;
517
518 for_each_zone(zone) {
519 struct per_cpu_pageset *pset;
520
521 pset = &zone->pageset[cpu];
522 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
523 struct per_cpu_pages *pcp;
524
525 pcp = &pset->pcp[i];
526 pcp->count -= free_pages_bulk(zone, pcp->count,
527 &pcp->list, 0);
528 }
529 }
530}
531#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
532
533#ifdef CONFIG_PM
534
535void mark_free_pages(struct zone *zone)
536{
537 unsigned long zone_pfn, flags;
538 int order;
539 struct list_head *curr;
540
541 if (!zone->spanned_pages)
542 return;
543
544 spin_lock_irqsave(&zone->lock, flags);
545 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
546 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
547
548 for (order = MAX_ORDER - 1; order >= 0; --order)
549 list_for_each(curr, &zone->free_area[order].free_list) {
550 unsigned long start_pfn, i;
551
552 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
553
554 for (i=0; i < (1<<order); i++)
555 SetPageNosaveFree(pfn_to_page(start_pfn+i));
556 }
557 spin_unlock_irqrestore(&zone->lock, flags);
558}
559
560/*
561 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
562 */
563void drain_local_pages(void)
564{
565 unsigned long flags;
566
567 local_irq_save(flags);
568 __drain_pages(smp_processor_id());
569 local_irq_restore(flags);
570}
571#endif /* CONFIG_PM */
572
573static void zone_statistics(struct zonelist *zonelist, struct zone *z)
574{
575#ifdef CONFIG_NUMA
576 unsigned long flags;
577 int cpu;
578 pg_data_t *pg = z->zone_pgdat;
579 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
580 struct per_cpu_pageset *p;
581
582 local_irq_save(flags);
583 cpu = smp_processor_id();
584 p = &z->pageset[cpu];
585 if (pg == orig) {
586 z->pageset[cpu].numa_hit++;
587 } else {
588 p->numa_miss++;
589 zonelist->zones[0]->pageset[cpu].numa_foreign++;
590 }
591 if (pg == NODE_DATA(numa_node_id()))
592 p->local_node++;
593 else
594 p->other_node++;
595 local_irq_restore(flags);
596#endif
597}
598
599/*
600 * Free a 0-order page
601 */
602static void FASTCALL(free_hot_cold_page(struct page *page, int cold));
603static void fastcall free_hot_cold_page(struct page *page, int cold)
604{
605 struct zone *zone = page_zone(page);
606 struct per_cpu_pages *pcp;
607 unsigned long flags;
608
609 arch_free_page(page, 0);
610
611 kernel_map_pages(page, 1, 0);
612 inc_page_state(pgfree);
613 if (PageAnon(page))
614 page->mapping = NULL;
615 free_pages_check(__FUNCTION__, page);
616 pcp = &zone->pageset[get_cpu()].pcp[cold];
617 local_irq_save(flags);
618 if (pcp->count >= pcp->high)
619 pcp->count -= free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
620 list_add(&page->lru, &pcp->list);
621 pcp->count++;
622 local_irq_restore(flags);
623 put_cpu();
624}
625
626void fastcall free_hot_page(struct page *page)
627{
628 free_hot_cold_page(page, 0);
629}
630
631void fastcall free_cold_page(struct page *page)
632{
633 free_hot_cold_page(page, 1);
634}
635
636static inline void prep_zero_page(struct page *page, int order, unsigned int __nocast gfp_flags)
637{
638 int i;
639
640 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
641 for(i = 0; i < (1 << order); i++)
642 clear_highpage(page + i);
643}
644
645/*
646 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
647 * we cheat by calling it from here, in the order > 0 path. Saves a branch
648 * or two.
649 */
650static struct page *
651buffered_rmqueue(struct zone *zone, int order, unsigned int __nocast gfp_flags)
652{
653 unsigned long flags;
654 struct page *page = NULL;
655 int cold = !!(gfp_flags & __GFP_COLD);
656
657 if (order == 0) {
658 struct per_cpu_pages *pcp;
659
660 pcp = &zone->pageset[get_cpu()].pcp[cold];
661 local_irq_save(flags);
662 if (pcp->count <= pcp->low)
663 pcp->count += rmqueue_bulk(zone, 0,
664 pcp->batch, &pcp->list);
665 if (pcp->count) {
666 page = list_entry(pcp->list.next, struct page, lru);
667 list_del(&page->lru);
668 pcp->count--;
669 }
670 local_irq_restore(flags);
671 put_cpu();
672 }
673
674 if (page == NULL) {
675 spin_lock_irqsave(&zone->lock, flags);
676 page = __rmqueue(zone, order);
677 spin_unlock_irqrestore(&zone->lock, flags);
678 }
679
680 if (page != NULL) {
681 BUG_ON(bad_range(zone, page));
682 mod_page_state_zone(zone, pgalloc, 1 << order);
683 prep_new_page(page, order);
684
685 if (gfp_flags & __GFP_ZERO)
686 prep_zero_page(page, order, gfp_flags);
687
688 if (order && (gfp_flags & __GFP_COMP))
689 prep_compound_page(page, order);
690 }
691 return page;
692}
693
694/*
695 * Return 1 if free pages are above 'mark'. This takes into account the order
696 * of the allocation.
697 */
698int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
699 int classzone_idx, int can_try_harder, int gfp_high)
700{
701 /* free_pages my go negative - that's OK */
702 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
703 int o;
704
705 if (gfp_high)
706 min -= min / 2;
707 if (can_try_harder)
708 min -= min / 4;
709
710 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
711 return 0;
712 for (o = 0; o < order; o++) {
713 /* At the next order, this order's pages become unavailable */
714 free_pages -= z->free_area[o].nr_free << o;
715
716 /* Require fewer higher order pages to be free */
717 min >>= 1;
718
719 if (free_pages <= min)
720 return 0;
721 }
722 return 1;
723}
724
725/*
726 * This is the 'heart' of the zoned buddy allocator.
727 */
728struct page * fastcall
729__alloc_pages(unsigned int __nocast gfp_mask, unsigned int order,
730 struct zonelist *zonelist)
731{
732 const int wait = gfp_mask & __GFP_WAIT;
733 struct zone **zones, *z;
734 struct page *page;
735 struct reclaim_state reclaim_state;
736 struct task_struct *p = current;
737 int i;
738 int classzone_idx;
739 int do_retry;
740 int can_try_harder;
741 int did_some_progress;
742
743 might_sleep_if(wait);
744
745 /*
746 * The caller may dip into page reserves a bit more if the caller
747 * cannot run direct reclaim, or is the caller has realtime scheduling
748 * policy
749 */
750 can_try_harder = (unlikely(rt_task(p)) && !in_interrupt()) || !wait;
751
752 zones = zonelist->zones; /* the list of zones suitable for gfp_mask */
753
754 if (unlikely(zones[0] == NULL)) {
755 /* Should this ever happen?? */
756 return NULL;
757 }
758
759 classzone_idx = zone_idx(zones[0]);
760
761 restart:
762 /* Go through the zonelist once, looking for a zone with enough free */
763 for (i = 0; (z = zones[i]) != NULL; i++) {
764
765 if (!zone_watermark_ok(z, order, z->pages_low,
766 classzone_idx, 0, 0))
767 continue;
768
769 if (!cpuset_zone_allowed(z))
770 continue;
771
772 page = buffered_rmqueue(z, order, gfp_mask);
773 if (page)
774 goto got_pg;
775 }
776
777 for (i = 0; (z = zones[i]) != NULL; i++)
778 wakeup_kswapd(z, order);
779
780 /*
781 * Go through the zonelist again. Let __GFP_HIGH and allocations
782 * coming from realtime tasks to go deeper into reserves
783 *
784 * This is the last chance, in general, before the goto nopage.
785 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
786 */
787 for (i = 0; (z = zones[i]) != NULL; i++) {
788 if (!zone_watermark_ok(z, order, z->pages_min,
789 classzone_idx, can_try_harder,
790 gfp_mask & __GFP_HIGH))
791 continue;
792
793 if (wait && !cpuset_zone_allowed(z))
794 continue;
795
796 page = buffered_rmqueue(z, order, gfp_mask);
797 if (page)
798 goto got_pg;
799 }
800
801 /* This allocation should allow future memory freeing. */
802 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) && !in_interrupt()) {
803 /* go through the zonelist yet again, ignoring mins */
804 for (i = 0; (z = zones[i]) != NULL; i++) {
805 if (!cpuset_zone_allowed(z))
806 continue;
807 page = buffered_rmqueue(z, order, gfp_mask);
808 if (page)
809 goto got_pg;
810 }
811 goto nopage;
812 }
813
814 /* Atomic allocations - we can't balance anything */
815 if (!wait)
816 goto nopage;
817
818rebalance:
819 cond_resched();
820
821 /* We now go into synchronous reclaim */
822 p->flags |= PF_MEMALLOC;
823 reclaim_state.reclaimed_slab = 0;
824 p->reclaim_state = &reclaim_state;
825
826 did_some_progress = try_to_free_pages(zones, gfp_mask, order);
827
828 p->reclaim_state = NULL;
829 p->flags &= ~PF_MEMALLOC;
830
831 cond_resched();
832
833 if (likely(did_some_progress)) {
834 /*
835 * Go through the zonelist yet one more time, keep
836 * very high watermark here, this is only to catch
837 * a parallel oom killing, we must fail if we're still
838 * under heavy pressure.
839 */
840 for (i = 0; (z = zones[i]) != NULL; i++) {
841 if (!zone_watermark_ok(z, order, z->pages_min,
842 classzone_idx, can_try_harder,
843 gfp_mask & __GFP_HIGH))
844 continue;
845
846 if (!cpuset_zone_allowed(z))
847 continue;
848
849 page = buffered_rmqueue(z, order, gfp_mask);
850 if (page)
851 goto got_pg;
852 }
853 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
854 /*
855 * Go through the zonelist yet one more time, keep
856 * very high watermark here, this is only to catch
857 * a parallel oom killing, we must fail if we're still
858 * under heavy pressure.
859 */
860 for (i = 0; (z = zones[i]) != NULL; i++) {
861 if (!zone_watermark_ok(z, order, z->pages_high,
862 classzone_idx, 0, 0))
863 continue;
864
865 if (!cpuset_zone_allowed(z))
866 continue;
867
868 page = buffered_rmqueue(z, order, gfp_mask);
869 if (page)
870 goto got_pg;
871 }
872
873 out_of_memory(gfp_mask);
874 goto restart;
875 }
876
877 /*
878 * Don't let big-order allocations loop unless the caller explicitly
879 * requests that. Wait for some write requests to complete then retry.
880 *
881 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
882 * <= 3, but that may not be true in other implementations.
883 */
884 do_retry = 0;
885 if (!(gfp_mask & __GFP_NORETRY)) {
886 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
887 do_retry = 1;
888 if (gfp_mask & __GFP_NOFAIL)
889 do_retry = 1;
890 }
891 if (do_retry) {
892 blk_congestion_wait(WRITE, HZ/50);
893 goto rebalance;
894 }
895
896nopage:
897 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
898 printk(KERN_WARNING "%s: page allocation failure."
899 " order:%d, mode:0x%x\n",
900 p->comm, order, gfp_mask);
901 dump_stack();
902 }
903 return NULL;
904got_pg:
905 zone_statistics(zonelist, z);
906 return page;
907}
908
909EXPORT_SYMBOL(__alloc_pages);
910
911/*
912 * Common helper functions.
913 */
914fastcall unsigned long __get_free_pages(unsigned int __nocast gfp_mask, unsigned int order)
915{
916 struct page * page;
917 page = alloc_pages(gfp_mask, order);
918 if (!page)
919 return 0;
920 return (unsigned long) page_address(page);
921}
922
923EXPORT_SYMBOL(__get_free_pages);
924
925fastcall unsigned long get_zeroed_page(unsigned int __nocast gfp_mask)
926{
927 struct page * page;
928
929 /*
930 * get_zeroed_page() returns a 32-bit address, which cannot represent
931 * a highmem page
932 */
933 BUG_ON(gfp_mask & __GFP_HIGHMEM);
934
935 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
936 if (page)
937 return (unsigned long) page_address(page);
938 return 0;
939}
940
941EXPORT_SYMBOL(get_zeroed_page);
942
943void __pagevec_free(struct pagevec *pvec)
944{
945 int i = pagevec_count(pvec);
946
947 while (--i >= 0)
948 free_hot_cold_page(pvec->pages[i], pvec->cold);
949}
950
951fastcall void __free_pages(struct page *page, unsigned int order)
952{
953 if (!PageReserved(page) && put_page_testzero(page)) {
954 if (order == 0)
955 free_hot_page(page);
956 else
957 __free_pages_ok(page, order);
958 }
959}
960
961EXPORT_SYMBOL(__free_pages);
962
963fastcall void free_pages(unsigned long addr, unsigned int order)
964{
965 if (addr != 0) {
966 BUG_ON(!virt_addr_valid((void *)addr));
967 __free_pages(virt_to_page((void *)addr), order);
968 }
969}
970
971EXPORT_SYMBOL(free_pages);
972
973/*
974 * Total amount of free (allocatable) RAM:
975 */
976unsigned int nr_free_pages(void)
977{
978 unsigned int sum = 0;
979 struct zone *zone;
980
981 for_each_zone(zone)
982 sum += zone->free_pages;
983
984 return sum;
985}
986
987EXPORT_SYMBOL(nr_free_pages);
988
989#ifdef CONFIG_NUMA
990unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
991{
992 unsigned int i, sum = 0;
993
994 for (i = 0; i < MAX_NR_ZONES; i++)
995 sum += pgdat->node_zones[i].free_pages;
996
997 return sum;
998}
999#endif
1000
1001static unsigned int nr_free_zone_pages(int offset)
1002{
1003 pg_data_t *pgdat;
1004 unsigned int sum = 0;
1005
1006 for_each_pgdat(pgdat) {
1007 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1008 struct zone **zonep = zonelist->zones;
1009 struct zone *zone;
1010
1011 for (zone = *zonep++; zone; zone = *zonep++) {
1012 unsigned long size = zone->present_pages;
1013 unsigned long high = zone->pages_high;
1014 if (size > high)
1015 sum += size - high;
1016 }
1017 }
1018
1019 return sum;
1020}
1021
1022/*
1023 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1024 */
1025unsigned int nr_free_buffer_pages(void)
1026{
1027 return nr_free_zone_pages(GFP_USER & GFP_ZONEMASK);
1028}
1029
1030/*
1031 * Amount of free RAM allocatable within all zones
1032 */
1033unsigned int nr_free_pagecache_pages(void)
1034{
1035 return nr_free_zone_pages(GFP_HIGHUSER & GFP_ZONEMASK);
1036}
1037
1038#ifdef CONFIG_HIGHMEM
1039unsigned int nr_free_highpages (void)
1040{
1041 pg_data_t *pgdat;
1042 unsigned int pages = 0;
1043
1044 for_each_pgdat(pgdat)
1045 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1046
1047 return pages;
1048}
1049#endif
1050
1051#ifdef CONFIG_NUMA
1052static void show_node(struct zone *zone)
1053{
1054 printk("Node %d ", zone->zone_pgdat->node_id);
1055}
1056#else
1057#define show_node(zone) do { } while (0)
1058#endif
1059
1060/*
1061 * Accumulate the page_state information across all CPUs.
1062 * The result is unavoidably approximate - it can change
1063 * during and after execution of this function.
1064 */
1065static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1066
1067atomic_t nr_pagecache = ATOMIC_INIT(0);
1068EXPORT_SYMBOL(nr_pagecache);
1069#ifdef CONFIG_SMP
1070DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1071#endif
1072
1073void __get_page_state(struct page_state *ret, int nr)
1074{
1075 int cpu = 0;
1076
1077 memset(ret, 0, sizeof(*ret));
1078
1079 cpu = first_cpu(cpu_online_map);
1080 while (cpu < NR_CPUS) {
1081 unsigned long *in, *out, off;
1082
1083 in = (unsigned long *)&per_cpu(page_states, cpu);
1084
1085 cpu = next_cpu(cpu, cpu_online_map);
1086
1087 if (cpu < NR_CPUS)
1088 prefetch(&per_cpu(page_states, cpu));
1089
1090 out = (unsigned long *)ret;
1091 for (off = 0; off < nr; off++)
1092 *out++ += *in++;
1093 }
1094}
1095
1096void get_page_state(struct page_state *ret)
1097{
1098 int nr;
1099
1100 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1101 nr /= sizeof(unsigned long);
1102
1103 __get_page_state(ret, nr + 1);
1104}
1105
1106void get_full_page_state(struct page_state *ret)
1107{
1108 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long));
1109}
1110
1111unsigned long __read_page_state(unsigned offset)
1112{
1113 unsigned long ret = 0;
1114 int cpu;
1115
1116 for_each_online_cpu(cpu) {
1117 unsigned long in;
1118
1119 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1120 ret += *((unsigned long *)in);
1121 }
1122 return ret;
1123}
1124
1125void __mod_page_state(unsigned offset, unsigned long delta)
1126{
1127 unsigned long flags;
1128 void* ptr;
1129
1130 local_irq_save(flags);
1131 ptr = &__get_cpu_var(page_states);
1132 *(unsigned long*)(ptr + offset) += delta;
1133 local_irq_restore(flags);
1134}
1135
1136EXPORT_SYMBOL(__mod_page_state);
1137
1138void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1139 unsigned long *free, struct pglist_data *pgdat)
1140{
1141 struct zone *zones = pgdat->node_zones;
1142 int i;
1143
1144 *active = 0;
1145 *inactive = 0;
1146 *free = 0;
1147 for (i = 0; i < MAX_NR_ZONES; i++) {
1148 *active += zones[i].nr_active;
1149 *inactive += zones[i].nr_inactive;
1150 *free += zones[i].free_pages;
1151 }
1152}
1153
1154void get_zone_counts(unsigned long *active,
1155 unsigned long *inactive, unsigned long *free)
1156{
1157 struct pglist_data *pgdat;
1158
1159 *active = 0;
1160 *inactive = 0;
1161 *free = 0;
1162 for_each_pgdat(pgdat) {
1163 unsigned long l, m, n;
1164 __get_zone_counts(&l, &m, &n, pgdat);
1165 *active += l;
1166 *inactive += m;
1167 *free += n;
1168 }
1169}
1170
1171void si_meminfo(struct sysinfo *val)
1172{
1173 val->totalram = totalram_pages;
1174 val->sharedram = 0;
1175 val->freeram = nr_free_pages();
1176 val->bufferram = nr_blockdev_pages();
1177#ifdef CONFIG_HIGHMEM
1178 val->totalhigh = totalhigh_pages;
1179 val->freehigh = nr_free_highpages();
1180#else
1181 val->totalhigh = 0;
1182 val->freehigh = 0;
1183#endif
1184 val->mem_unit = PAGE_SIZE;
1185}
1186
1187EXPORT_SYMBOL(si_meminfo);
1188
1189#ifdef CONFIG_NUMA
1190void si_meminfo_node(struct sysinfo *val, int nid)
1191{
1192 pg_data_t *pgdat = NODE_DATA(nid);
1193
1194 val->totalram = pgdat->node_present_pages;
1195 val->freeram = nr_free_pages_pgdat(pgdat);
1196 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1197 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1198 val->mem_unit = PAGE_SIZE;
1199}
1200#endif
1201
1202#define K(x) ((x) << (PAGE_SHIFT-10))
1203
1204/*
1205 * Show free area list (used inside shift_scroll-lock stuff)
1206 * We also calculate the percentage fragmentation. We do this by counting the
1207 * memory on each free list with the exception of the first item on the list.
1208 */
1209void show_free_areas(void)
1210{
1211 struct page_state ps;
1212 int cpu, temperature;
1213 unsigned long active;
1214 unsigned long inactive;
1215 unsigned long free;
1216 struct zone *zone;
1217
1218 for_each_zone(zone) {
1219 show_node(zone);
1220 printk("%s per-cpu:", zone->name);
1221
1222 if (!zone->present_pages) {
1223 printk(" empty\n");
1224 continue;
1225 } else
1226 printk("\n");
1227
1228 for (cpu = 0; cpu < NR_CPUS; ++cpu) {
1229 struct per_cpu_pageset *pageset;
1230
1231 if (!cpu_possible(cpu))
1232 continue;
1233
1234 pageset = zone->pageset + cpu;
1235
1236 for (temperature = 0; temperature < 2; temperature++)
1237 printk("cpu %d %s: low %d, high %d, batch %d\n",
1238 cpu,
1239 temperature ? "cold" : "hot",
1240 pageset->pcp[temperature].low,
1241 pageset->pcp[temperature].high,
1242 pageset->pcp[temperature].batch);
1243 }
1244 }
1245
1246 get_page_state(&ps);
1247 get_zone_counts(&active, &inactive, &free);
1248
1249 printk("\nFree pages: %11ukB (%ukB HighMem)\n",
1250 K(nr_free_pages()),
1251 K(nr_free_highpages()));
1252
1253 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1254 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1255 active,
1256 inactive,
1257 ps.nr_dirty,
1258 ps.nr_writeback,
1259 ps.nr_unstable,
1260 nr_free_pages(),
1261 ps.nr_slab,
1262 ps.nr_mapped,
1263 ps.nr_page_table_pages);
1264
1265 for_each_zone(zone) {
1266 int i;
1267
1268 show_node(zone);
1269 printk("%s"
1270 " free:%lukB"
1271 " min:%lukB"
1272 " low:%lukB"
1273 " high:%lukB"
1274 " active:%lukB"
1275 " inactive:%lukB"
1276 " present:%lukB"
1277 " pages_scanned:%lu"
1278 " all_unreclaimable? %s"
1279 "\n",
1280 zone->name,
1281 K(zone->free_pages),
1282 K(zone->pages_min),
1283 K(zone->pages_low),
1284 K(zone->pages_high),
1285 K(zone->nr_active),
1286 K(zone->nr_inactive),
1287 K(zone->present_pages),
1288 zone->pages_scanned,
1289 (zone->all_unreclaimable ? "yes" : "no")
1290 );
1291 printk("lowmem_reserve[]:");
1292 for (i = 0; i < MAX_NR_ZONES; i++)
1293 printk(" %lu", zone->lowmem_reserve[i]);
1294 printk("\n");
1295 }
1296
1297 for_each_zone(zone) {
1298 unsigned long nr, flags, order, total = 0;
1299
1300 show_node(zone);
1301 printk("%s: ", zone->name);
1302 if (!zone->present_pages) {
1303 printk("empty\n");
1304 continue;
1305 }
1306
1307 spin_lock_irqsave(&zone->lock, flags);
1308 for (order = 0; order < MAX_ORDER; order++) {
1309 nr = zone->free_area[order].nr_free;
1310 total += nr << order;
1311 printk("%lu*%lukB ", nr, K(1UL) << order);
1312 }
1313 spin_unlock_irqrestore(&zone->lock, flags);
1314 printk("= %lukB\n", K(total));
1315 }
1316
1317 show_swap_cache_info();
1318}
1319
1320/*
1321 * Builds allocation fallback zone lists.
1322 */
1323static int __init build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, int j, int k)
1324{
1325 switch (k) {
1326 struct zone *zone;
1327 default:
1328 BUG();
1329 case ZONE_HIGHMEM:
1330 zone = pgdat->node_zones + ZONE_HIGHMEM;
1331 if (zone->present_pages) {
1332#ifndef CONFIG_HIGHMEM
1333 BUG();
1334#endif
1335 zonelist->zones[j++] = zone;
1336 }
1337 case ZONE_NORMAL:
1338 zone = pgdat->node_zones + ZONE_NORMAL;
1339 if (zone->present_pages)
1340 zonelist->zones[j++] = zone;
1341 case ZONE_DMA:
1342 zone = pgdat->node_zones + ZONE_DMA;
1343 if (zone->present_pages)
1344 zonelist->zones[j++] = zone;
1345 }
1346
1347 return j;
1348}
1349
1350#ifdef CONFIG_NUMA
1351#define MAX_NODE_LOAD (num_online_nodes())
1352static int __initdata node_load[MAX_NUMNODES];
1353/**
1354 * find_next_best_node - find the next node that should appear in a given
1355 * node's fallback list
1356 * @node: node whose fallback list we're appending
1357 * @used_node_mask: nodemask_t of already used nodes
1358 *
1359 * We use a number of factors to determine which is the next node that should
1360 * appear on a given node's fallback list. The node should not have appeared
1361 * already in @node's fallback list, and it should be the next closest node
1362 * according to the distance array (which contains arbitrary distance values
1363 * from each node to each node in the system), and should also prefer nodes
1364 * with no CPUs, since presumably they'll have very little allocation pressure
1365 * on them otherwise.
1366 * It returns -1 if no node is found.
1367 */
1368static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1369{
1370 int i, n, val;
1371 int min_val = INT_MAX;
1372 int best_node = -1;
1373
1374 for_each_online_node(i) {
1375 cpumask_t tmp;
1376
1377 /* Start from local node */
1378 n = (node+i) % num_online_nodes();
1379
1380 /* Don't want a node to appear more than once */
1381 if (node_isset(n, *used_node_mask))
1382 continue;
1383
1384 /* Use the local node if we haven't already */
1385 if (!node_isset(node, *used_node_mask)) {
1386 best_node = node;
1387 break;
1388 }
1389
1390 /* Use the distance array to find the distance */
1391 val = node_distance(node, n);
1392
1393 /* Give preference to headless and unused nodes */
1394 tmp = node_to_cpumask(n);
1395 if (!cpus_empty(tmp))
1396 val += PENALTY_FOR_NODE_WITH_CPUS;
1397
1398 /* Slight preference for less loaded node */
1399 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1400 val += node_load[n];
1401
1402 if (val < min_val) {
1403 min_val = val;
1404 best_node = n;
1405 }
1406 }
1407
1408 if (best_node >= 0)
1409 node_set(best_node, *used_node_mask);
1410
1411 return best_node;
1412}
1413
1414static void __init build_zonelists(pg_data_t *pgdat)
1415{
1416 int i, j, k, node, local_node;
1417 int prev_node, load;
1418 struct zonelist *zonelist;
1419 nodemask_t used_mask;
1420
1421 /* initialize zonelists */
1422 for (i = 0; i < GFP_ZONETYPES; i++) {
1423 zonelist = pgdat->node_zonelists + i;
1424 zonelist->zones[0] = NULL;
1425 }
1426
1427 /* NUMA-aware ordering of nodes */
1428 local_node = pgdat->node_id;
1429 load = num_online_nodes();
1430 prev_node = local_node;
1431 nodes_clear(used_mask);
1432 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1433 /*
1434 * We don't want to pressure a particular node.
1435 * So adding penalty to the first node in same
1436 * distance group to make it round-robin.
1437 */
1438 if (node_distance(local_node, node) !=
1439 node_distance(local_node, prev_node))
1440 node_load[node] += load;
1441 prev_node = node;
1442 load--;
1443 for (i = 0; i < GFP_ZONETYPES; i++) {
1444 zonelist = pgdat->node_zonelists + i;
1445 for (j = 0; zonelist->zones[j] != NULL; j++);
1446
1447 k = ZONE_NORMAL;
1448 if (i & __GFP_HIGHMEM)
1449 k = ZONE_HIGHMEM;
1450 if (i & __GFP_DMA)
1451 k = ZONE_DMA;
1452
1453 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1454 zonelist->zones[j] = NULL;
1455 }
1456 }
1457}
1458
1459#else /* CONFIG_NUMA */
1460
1461static void __init build_zonelists(pg_data_t *pgdat)
1462{
1463 int i, j, k, node, local_node;
1464
1465 local_node = pgdat->node_id;
1466 for (i = 0; i < GFP_ZONETYPES; i++) {
1467 struct zonelist *zonelist;
1468
1469 zonelist = pgdat->node_zonelists + i;
1470
1471 j = 0;
1472 k = ZONE_NORMAL;
1473 if (i & __GFP_HIGHMEM)
1474 k = ZONE_HIGHMEM;
1475 if (i & __GFP_DMA)
1476 k = ZONE_DMA;
1477
1478 j = build_zonelists_node(pgdat, zonelist, j, k);
1479 /*
1480 * Now we build the zonelist so that it contains the zones
1481 * of all the other nodes.
1482 * We don't want to pressure a particular node, so when
1483 * building the zones for node N, we make sure that the
1484 * zones coming right after the local ones are those from
1485 * node N+1 (modulo N)
1486 */
1487 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1488 if (!node_online(node))
1489 continue;
1490 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1491 }
1492 for (node = 0; node < local_node; node++) {
1493 if (!node_online(node))
1494 continue;
1495 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1496 }
1497
1498 zonelist->zones[j] = NULL;
1499 }
1500}
1501
1502#endif /* CONFIG_NUMA */
1503
1504void __init build_all_zonelists(void)
1505{
1506 int i;
1507
1508 for_each_online_node(i)
1509 build_zonelists(NODE_DATA(i));
1510 printk("Built %i zonelists\n", num_online_nodes());
1511 cpuset_init_current_mems_allowed();
1512}
1513
1514/*
1515 * Helper functions to size the waitqueue hash table.
1516 * Essentially these want to choose hash table sizes sufficiently
1517 * large so that collisions trying to wait on pages are rare.
1518 * But in fact, the number of active page waitqueues on typical
1519 * systems is ridiculously low, less than 200. So this is even
1520 * conservative, even though it seems large.
1521 *
1522 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1523 * waitqueues, i.e. the size of the waitq table given the number of pages.
1524 */
1525#define PAGES_PER_WAITQUEUE 256
1526
1527static inline unsigned long wait_table_size(unsigned long pages)
1528{
1529 unsigned long size = 1;
1530
1531 pages /= PAGES_PER_WAITQUEUE;
1532
1533 while (size < pages)
1534 size <<= 1;
1535
1536 /*
1537 * Once we have dozens or even hundreds of threads sleeping
1538 * on IO we've got bigger problems than wait queue collision.
1539 * Limit the size of the wait table to a reasonable size.
1540 */
1541 size = min(size, 4096UL);
1542
1543 return max(size, 4UL);
1544}
1545
1546/*
1547 * This is an integer logarithm so that shifts can be used later
1548 * to extract the more random high bits from the multiplicative
1549 * hash function before the remainder is taken.
1550 */
1551static inline unsigned long wait_table_bits(unsigned long size)
1552{
1553 return ffz(~size);
1554}
1555
1556#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1557
1558static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1559 unsigned long *zones_size, unsigned long *zholes_size)
1560{
1561 unsigned long realtotalpages, totalpages = 0;
1562 int i;
1563
1564 for (i = 0; i < MAX_NR_ZONES; i++)
1565 totalpages += zones_size[i];
1566 pgdat->node_spanned_pages = totalpages;
1567
1568 realtotalpages = totalpages;
1569 if (zholes_size)
1570 for (i = 0; i < MAX_NR_ZONES; i++)
1571 realtotalpages -= zholes_size[i];
1572 pgdat->node_present_pages = realtotalpages;
1573 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1574}
1575
1576
1577/*
1578 * Initially all pages are reserved - free ones are freed
1579 * up by free_all_bootmem() once the early boot process is
1580 * done. Non-atomic initialization, single-pass.
1581 */
1582void __init memmap_init_zone(unsigned long size, int nid, unsigned long zone,
1583 unsigned long start_pfn)
1584{
1585 struct page *start = pfn_to_page(start_pfn);
1586 struct page *page;
1587
1588 for (page = start; page < (start + size); page++) {
1589 set_page_zone(page, NODEZONE(nid, zone));
1590 set_page_count(page, 0);
1591 reset_page_mapcount(page);
1592 SetPageReserved(page);
1593 INIT_LIST_HEAD(&page->lru);
1594#ifdef WANT_PAGE_VIRTUAL
1595 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1596 if (!is_highmem_idx(zone))
1597 set_page_address(page, __va(start_pfn << PAGE_SHIFT));
1598#endif
1599 start_pfn++;
1600 }
1601}
1602
1603void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1604 unsigned long size)
1605{
1606 int order;
1607 for (order = 0; order < MAX_ORDER ; order++) {
1608 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1609 zone->free_area[order].nr_free = 0;
1610 }
1611}
1612
1613#ifndef __HAVE_ARCH_MEMMAP_INIT
1614#define memmap_init(size, nid, zone, start_pfn) \
1615 memmap_init_zone((size), (nid), (zone), (start_pfn))
1616#endif
1617
1618/*
1619 * Set up the zone data structures:
1620 * - mark all pages reserved
1621 * - mark all memory queues empty
1622 * - clear the memory bitmaps
1623 */
1624static void __init free_area_init_core(struct pglist_data *pgdat,
1625 unsigned long *zones_size, unsigned long *zholes_size)
1626{
1627 unsigned long i, j;
1628 const unsigned long zone_required_alignment = 1UL << (MAX_ORDER-1);
1629 int cpu, nid = pgdat->node_id;
1630 unsigned long zone_start_pfn = pgdat->node_start_pfn;
1631
1632 pgdat->nr_zones = 0;
1633 init_waitqueue_head(&pgdat->kswapd_wait);
1634 pgdat->kswapd_max_order = 0;
1635
1636 for (j = 0; j < MAX_NR_ZONES; j++) {
1637 struct zone *zone = pgdat->node_zones + j;
1638 unsigned long size, realsize;
1639 unsigned long batch;
1640
1641 zone_table[NODEZONE(nid, j)] = zone;
1642 realsize = size = zones_size[j];
1643 if (zholes_size)
1644 realsize -= zholes_size[j];
1645
1646 if (j == ZONE_DMA || j == ZONE_NORMAL)
1647 nr_kernel_pages += realsize;
1648 nr_all_pages += realsize;
1649
1650 zone->spanned_pages = size;
1651 zone->present_pages = realsize;
1652 zone->name = zone_names[j];
1653 spin_lock_init(&zone->lock);
1654 spin_lock_init(&zone->lru_lock);
1655 zone->zone_pgdat = pgdat;
1656 zone->free_pages = 0;
1657
1658 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
1659
1660 /*
1661 * The per-cpu-pages pools are set to around 1000th of the
1662 * size of the zone. But no more than 1/4 of a meg - there's
1663 * no point in going beyond the size of L2 cache.
1664 *
1665 * OK, so we don't know how big the cache is. So guess.
1666 */
1667 batch = zone->present_pages / 1024;
1668 if (batch * PAGE_SIZE > 256 * 1024)
1669 batch = (256 * 1024) / PAGE_SIZE;
1670 batch /= 4; /* We effectively *= 4 below */
1671 if (batch < 1)
1672 batch = 1;
1673
1674 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1675 struct per_cpu_pages *pcp;
1676
1677 pcp = &zone->pageset[cpu].pcp[0]; /* hot */
1678 pcp->count = 0;
1679 pcp->low = 2 * batch;
1680 pcp->high = 6 * batch;
1681 pcp->batch = 1 * batch;
1682 INIT_LIST_HEAD(&pcp->list);
1683
1684 pcp = &zone->pageset[cpu].pcp[1]; /* cold */
1685 pcp->count = 0;
1686 pcp->low = 0;
1687 pcp->high = 2 * batch;
1688 pcp->batch = 1 * batch;
1689 INIT_LIST_HEAD(&pcp->list);
1690 }
1691 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
1692 zone_names[j], realsize, batch);
1693 INIT_LIST_HEAD(&zone->active_list);
1694 INIT_LIST_HEAD(&zone->inactive_list);
1695 zone->nr_scan_active = 0;
1696 zone->nr_scan_inactive = 0;
1697 zone->nr_active = 0;
1698 zone->nr_inactive = 0;
1699 if (!size)
1700 continue;
1701
1702 /*
1703 * The per-page waitqueue mechanism uses hashed waitqueues
1704 * per zone.
1705 */
1706 zone->wait_table_size = wait_table_size(size);
1707 zone->wait_table_bits =
1708 wait_table_bits(zone->wait_table_size);
1709 zone->wait_table = (wait_queue_head_t *)
1710 alloc_bootmem_node(pgdat, zone->wait_table_size
1711 * sizeof(wait_queue_head_t));
1712
1713 for(i = 0; i < zone->wait_table_size; ++i)
1714 init_waitqueue_head(zone->wait_table + i);
1715
1716 pgdat->nr_zones = j+1;
1717
1718 zone->zone_mem_map = pfn_to_page(zone_start_pfn);
1719 zone->zone_start_pfn = zone_start_pfn;
1720
1721 if ((zone_start_pfn) & (zone_required_alignment-1))
1722 printk(KERN_CRIT "BUG: wrong zone alignment, it will crash\n");
1723
1724 memmap_init(size, nid, j, zone_start_pfn);
1725
1726 zone_start_pfn += size;
1727
1728 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
1729 }
1730}
1731
1732static void __init alloc_node_mem_map(struct pglist_data *pgdat)
1733{
1734 unsigned long size;
1735
1736 /* Skip empty nodes */
1737 if (!pgdat->node_spanned_pages)
1738 return;
1739
1740 /* ia64 gets its own node_mem_map, before this, without bootmem */
1741 if (!pgdat->node_mem_map) {
1742 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
1743 pgdat->node_mem_map = alloc_bootmem_node(pgdat, size);
1744 }
1745#ifndef CONFIG_DISCONTIGMEM
1746 /*
1747 * With no DISCONTIG, the global mem_map is just set as node 0's
1748 */
1749 if (pgdat == NODE_DATA(0))
1750 mem_map = NODE_DATA(0)->node_mem_map;
1751#endif
1752}
1753
1754void __init free_area_init_node(int nid, struct pglist_data *pgdat,
1755 unsigned long *zones_size, unsigned long node_start_pfn,
1756 unsigned long *zholes_size)
1757{
1758 pgdat->node_id = nid;
1759 pgdat->node_start_pfn = node_start_pfn;
1760 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
1761
1762 alloc_node_mem_map(pgdat);
1763
1764 free_area_init_core(pgdat, zones_size, zholes_size);
1765}
1766
1767#ifndef CONFIG_DISCONTIGMEM
1768static bootmem_data_t contig_bootmem_data;
1769struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
1770
1771EXPORT_SYMBOL(contig_page_data);
1772
1773void __init free_area_init(unsigned long *zones_size)
1774{
1775 free_area_init_node(0, &contig_page_data, zones_size,
1776 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
1777}
1778#endif
1779
1780#ifdef CONFIG_PROC_FS
1781
1782#include <linux/seq_file.h>
1783
1784static void *frag_start(struct seq_file *m, loff_t *pos)
1785{
1786 pg_data_t *pgdat;
1787 loff_t node = *pos;
1788
1789 for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next)
1790 --node;
1791
1792 return pgdat;
1793}
1794
1795static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
1796{
1797 pg_data_t *pgdat = (pg_data_t *)arg;
1798
1799 (*pos)++;
1800 return pgdat->pgdat_next;
1801}
1802
1803static void frag_stop(struct seq_file *m, void *arg)
1804{
1805}
1806
1807/*
1808 * This walks the free areas for each zone.
1809 */
1810static int frag_show(struct seq_file *m, void *arg)
1811{
1812 pg_data_t *pgdat = (pg_data_t *)arg;
1813 struct zone *zone;
1814 struct zone *node_zones = pgdat->node_zones;
1815 unsigned long flags;
1816 int order;
1817
1818 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
1819 if (!zone->present_pages)
1820 continue;
1821
1822 spin_lock_irqsave(&zone->lock, flags);
1823 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
1824 for (order = 0; order < MAX_ORDER; ++order)
1825 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
1826 spin_unlock_irqrestore(&zone->lock, flags);
1827 seq_putc(m, '\n');
1828 }
1829 return 0;
1830}
1831
1832struct seq_operations fragmentation_op = {
1833 .start = frag_start,
1834 .next = frag_next,
1835 .stop = frag_stop,
1836 .show = frag_show,
1837};
1838
1839static char *vmstat_text[] = {
1840 "nr_dirty",
1841 "nr_writeback",
1842 "nr_unstable",
1843 "nr_page_table_pages",
1844 "nr_mapped",
1845 "nr_slab",
1846
1847 "pgpgin",
1848 "pgpgout",
1849 "pswpin",
1850 "pswpout",
1851 "pgalloc_high",
1852
1853 "pgalloc_normal",
1854 "pgalloc_dma",
1855 "pgfree",
1856 "pgactivate",
1857 "pgdeactivate",
1858
1859 "pgfault",
1860 "pgmajfault",
1861 "pgrefill_high",
1862 "pgrefill_normal",
1863 "pgrefill_dma",
1864
1865 "pgsteal_high",
1866 "pgsteal_normal",
1867 "pgsteal_dma",
1868 "pgscan_kswapd_high",
1869 "pgscan_kswapd_normal",
1870
1871 "pgscan_kswapd_dma",
1872 "pgscan_direct_high",
1873 "pgscan_direct_normal",
1874 "pgscan_direct_dma",
1875 "pginodesteal",
1876
1877 "slabs_scanned",
1878 "kswapd_steal",
1879 "kswapd_inodesteal",
1880 "pageoutrun",
1881 "allocstall",
1882
1883 "pgrotated",
1884};
1885
1886static void *vmstat_start(struct seq_file *m, loff_t *pos)
1887{
1888 struct page_state *ps;
1889
1890 if (*pos >= ARRAY_SIZE(vmstat_text))
1891 return NULL;
1892
1893 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
1894 m->private = ps;
1895 if (!ps)
1896 return ERR_PTR(-ENOMEM);
1897 get_full_page_state(ps);
1898 ps->pgpgin /= 2; /* sectors -> kbytes */
1899 ps->pgpgout /= 2;
1900 return (unsigned long *)ps + *pos;
1901}
1902
1903static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
1904{
1905 (*pos)++;
1906 if (*pos >= ARRAY_SIZE(vmstat_text))
1907 return NULL;
1908 return (unsigned long *)m->private + *pos;
1909}
1910
1911static int vmstat_show(struct seq_file *m, void *arg)
1912{
1913 unsigned long *l = arg;
1914 unsigned long off = l - (unsigned long *)m->private;
1915
1916 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
1917 return 0;
1918}
1919
1920static void vmstat_stop(struct seq_file *m, void *arg)
1921{
1922 kfree(m->private);
1923 m->private = NULL;
1924}
1925
1926struct seq_operations vmstat_op = {
1927 .start = vmstat_start,
1928 .next = vmstat_next,
1929 .stop = vmstat_stop,
1930 .show = vmstat_show,
1931};
1932
1933#endif /* CONFIG_PROC_FS */
1934
1935#ifdef CONFIG_HOTPLUG_CPU
1936static int page_alloc_cpu_notify(struct notifier_block *self,
1937 unsigned long action, void *hcpu)
1938{
1939 int cpu = (unsigned long)hcpu;
1940 long *count;
1941 unsigned long *src, *dest;
1942
1943 if (action == CPU_DEAD) {
1944 int i;
1945
1946 /* Drain local pagecache count. */
1947 count = &per_cpu(nr_pagecache_local, cpu);
1948 atomic_add(*count, &nr_pagecache);
1949 *count = 0;
1950 local_irq_disable();
1951 __drain_pages(cpu);
1952
1953 /* Add dead cpu's page_states to our own. */
1954 dest = (unsigned long *)&__get_cpu_var(page_states);
1955 src = (unsigned long *)&per_cpu(page_states, cpu);
1956
1957 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
1958 i++) {
1959 dest[i] += src[i];
1960 src[i] = 0;
1961 }
1962
1963 local_irq_enable();
1964 }
1965 return NOTIFY_OK;
1966}
1967#endif /* CONFIG_HOTPLUG_CPU */
1968
1969void __init page_alloc_init(void)
1970{
1971 hotcpu_notifier(page_alloc_cpu_notify, 0);
1972}
1973
1974/*
1975 * setup_per_zone_lowmem_reserve - called whenever
1976 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
1977 * has a correct pages reserved value, so an adequate number of
1978 * pages are left in the zone after a successful __alloc_pages().
1979 */
1980static void setup_per_zone_lowmem_reserve(void)
1981{
1982 struct pglist_data *pgdat;
1983 int j, idx;
1984
1985 for_each_pgdat(pgdat) {
1986 for (j = 0; j < MAX_NR_ZONES; j++) {
1987 struct zone *zone = pgdat->node_zones + j;
1988 unsigned long present_pages = zone->present_pages;
1989
1990 zone->lowmem_reserve[j] = 0;
1991
1992 for (idx = j-1; idx >= 0; idx--) {
1993 struct zone *lower_zone;
1994
1995 if (sysctl_lowmem_reserve_ratio[idx] < 1)
1996 sysctl_lowmem_reserve_ratio[idx] = 1;
1997
1998 lower_zone = pgdat->node_zones + idx;
1999 lower_zone->lowmem_reserve[j] = present_pages /
2000 sysctl_lowmem_reserve_ratio[idx];
2001 present_pages += lower_zone->present_pages;
2002 }
2003 }
2004 }
2005}
2006
2007/*
2008 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2009 * that the pages_{min,low,high} values for each zone are set correctly
2010 * with respect to min_free_kbytes.
2011 */
2012static void setup_per_zone_pages_min(void)
2013{
2014 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2015 unsigned long lowmem_pages = 0;
2016 struct zone *zone;
2017 unsigned long flags;
2018
2019 /* Calculate total number of !ZONE_HIGHMEM pages */
2020 for_each_zone(zone) {
2021 if (!is_highmem(zone))
2022 lowmem_pages += zone->present_pages;
2023 }
2024
2025 for_each_zone(zone) {
2026 spin_lock_irqsave(&zone->lru_lock, flags);
2027 if (is_highmem(zone)) {
2028 /*
2029 * Often, highmem doesn't need to reserve any pages.
2030 * But the pages_min/low/high values are also used for
2031 * batching up page reclaim activity so we need a
2032 * decent value here.
2033 */
2034 int min_pages;
2035
2036 min_pages = zone->present_pages / 1024;
2037 if (min_pages < SWAP_CLUSTER_MAX)
2038 min_pages = SWAP_CLUSTER_MAX;
2039 if (min_pages > 128)
2040 min_pages = 128;
2041 zone->pages_min = min_pages;
2042 } else {
2043 /* if it's a lowmem zone, reserve a number of pages
2044 * proportionate to the zone's size.
2045 */
2046 zone->pages_min = (pages_min * zone->present_pages) /
2047 lowmem_pages;
2048 }
2049
2050 /*
2051 * When interpreting these watermarks, just keep in mind that:
2052 * zone->pages_min == (zone->pages_min * 4) / 4;
2053 */
2054 zone->pages_low = (zone->pages_min * 5) / 4;
2055 zone->pages_high = (zone->pages_min * 6) / 4;
2056 spin_unlock_irqrestore(&zone->lru_lock, flags);
2057 }
2058}
2059
2060/*
2061 * Initialise min_free_kbytes.
2062 *
2063 * For small machines we want it small (128k min). For large machines
2064 * we want it large (64MB max). But it is not linear, because network
2065 * bandwidth does not increase linearly with machine size. We use
2066 *
2067 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2068 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2069 *
2070 * which yields
2071 *
2072 * 16MB: 512k
2073 * 32MB: 724k
2074 * 64MB: 1024k
2075 * 128MB: 1448k
2076 * 256MB: 2048k
2077 * 512MB: 2896k
2078 * 1024MB: 4096k
2079 * 2048MB: 5792k
2080 * 4096MB: 8192k
2081 * 8192MB: 11584k
2082 * 16384MB: 16384k
2083 */
2084static int __init init_per_zone_pages_min(void)
2085{
2086 unsigned long lowmem_kbytes;
2087
2088 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2089
2090 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2091 if (min_free_kbytes < 128)
2092 min_free_kbytes = 128;
2093 if (min_free_kbytes > 65536)
2094 min_free_kbytes = 65536;
2095 setup_per_zone_pages_min();
2096 setup_per_zone_lowmem_reserve();
2097 return 0;
2098}
2099module_init(init_per_zone_pages_min)
2100
2101/*
2102 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2103 * that we can call two helper functions whenever min_free_kbytes
2104 * changes.
2105 */
2106int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2107 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2108{
2109 proc_dointvec(table, write, file, buffer, length, ppos);
2110 setup_per_zone_pages_min();
2111 return 0;
2112}
2113
2114/*
2115 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2116 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2117 * whenever sysctl_lowmem_reserve_ratio changes.
2118 *
2119 * The reserve ratio obviously has absolutely no relation with the
2120 * pages_min watermarks. The lowmem reserve ratio can only make sense
2121 * if in function of the boot time zone sizes.
2122 */
2123int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2124 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2125{
2126 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2127 setup_per_zone_lowmem_reserve();
2128 return 0;
2129}
2130
2131__initdata int hashdist = HASHDIST_DEFAULT;
2132
2133#ifdef CONFIG_NUMA
2134static int __init set_hashdist(char *str)
2135{
2136 if (!str)
2137 return 0;
2138 hashdist = simple_strtoul(str, &str, 0);
2139 return 1;
2140}
2141__setup("hashdist=", set_hashdist);
2142#endif
2143
2144/*
2145 * allocate a large system hash table from bootmem
2146 * - it is assumed that the hash table must contain an exact power-of-2
2147 * quantity of entries
2148 * - limit is the number of hash buckets, not the total allocation size
2149 */
2150void *__init alloc_large_system_hash(const char *tablename,
2151 unsigned long bucketsize,
2152 unsigned long numentries,
2153 int scale,
2154 int flags,
2155 unsigned int *_hash_shift,
2156 unsigned int *_hash_mask,
2157 unsigned long limit)
2158{
2159 unsigned long long max = limit;
2160 unsigned long log2qty, size;
2161 void *table = NULL;
2162
2163 /* allow the kernel cmdline to have a say */
2164 if (!numentries) {
2165 /* round applicable memory size up to nearest megabyte */
2166 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2167 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2168 numentries >>= 20 - PAGE_SHIFT;
2169 numentries <<= 20 - PAGE_SHIFT;
2170
2171 /* limit to 1 bucket per 2^scale bytes of low memory */
2172 if (scale > PAGE_SHIFT)
2173 numentries >>= (scale - PAGE_SHIFT);
2174 else
2175 numentries <<= (PAGE_SHIFT - scale);
2176 }
2177 /* rounded up to nearest power of 2 in size */
2178 numentries = 1UL << (long_log2(numentries) + 1);
2179
2180 /* limit allocation size to 1/16 total memory by default */
2181 if (max == 0) {
2182 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2183 do_div(max, bucketsize);
2184 }
2185
2186 if (numentries > max)
2187 numentries = max;
2188
2189 log2qty = long_log2(numentries);
2190
2191 do {
2192 size = bucketsize << log2qty;
2193 if (flags & HASH_EARLY)
2194 table = alloc_bootmem(size);
2195 else if (hashdist)
2196 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2197 else {
2198 unsigned long order;
2199 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2200 ;
2201 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2202 }
2203 } while (!table && size > PAGE_SIZE && --log2qty);
2204
2205 if (!table)
2206 panic("Failed to allocate %s hash table\n", tablename);
2207
2208 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2209 tablename,
2210 (1U << log2qty),
2211 long_log2(size) - PAGE_SHIFT,
2212 size);
2213
2214 if (_hash_shift)
2215 *_hash_shift = log2qty;
2216 if (_hash_mask)
2217 *_hash_mask = (1 << log2qty) - 1;
2218
2219 return table;
2220}